Thursday, October 22, 2015

New T-SQL programming enhancements in SQL Server 2012

In SQL Server 2012 and 2014 Microsoft introduced many new features and enhancements. In fact there were 14 new in-built functions introduced in SQL Server 2012.  However I noticed that for some reason I don’t see many people using them.  In my company we upgraded all our servers to 2014 from 2008R2 almost a year back but I see none of the developers using them or even aware.  Also in last 6 months I interviewed many people and one of the question I asked to candidates that are you aware of any new features of SQL Server 2012/14?  While most of the people talked about the changes in SSIS, project deployment etc but very few knew about the 14 new in-built functions introduced in SQL Server 2012. 

So I thought of writing about these new in-built functions. In this post I’m not going to write the details of these function.  This is just going to be a reminder that now we have 14 more in-built functions that can help us in writing SQL code. 

Conversion functions

1. PARSE (Transact-SQL) : PARSE function converts the string expression to the requested data type. And if it fails to translate then raises an exception.

2. TRY_CONVERT (Transact-SQL): Returns a value cast to the specified data type if the cast succeeds; otherwise, returns null.

3. TRY_PARSE (Transact-SQL): Returns the result of an expression, translated to the requested data type, or null if the cast fails

You might wonder what is the difference between PARSE and CONVERT or may be PARSE and TRY_PARSE; I’ll try to explain that in my next few posts.

 

Date and time functions

4. DATEFROMPARTS (Transact-SQL): Returns a date value for the specified year, month, and day.

5. DATETIME2FROMPARTS (Transact-SQL): Returns a datetime2 value for the specified date and time and with the specified precision.

6. DATETIMEFROMPARTS (Transact-SQL): Returns a datetime value for the specified date and time.

7. DATETIMEOFFSETFROMPARTS (Transact-SQL): Returns a datetimeoffset value for the specified date and time and with the specified offsets and precision.

8. EOMONTH (Transact-SQL): Returns the last day of the month that contains the specified date, with an optional offset.

9. SMALLDATETIMEFROMPARTS (Transact-SQL): Returns a smalldatetime value for the specified date and time.

10. TIMEFROMPARTS (Transact-SQL): Returns a time value for the specified time and with the specified precision.

 

Logical functions

11. CHOOSE (Transact-SQL): Returns the item at the specified index from a list of values.

12. IIF (Transact-SQL): Returns one of two values, depending on whether the Boolean expression evaluates to true or false.

 

String functions

13. CONCAT (Transact-SQL): Returns a string that is the result of concatenating two or more string values.

14. FORMAT (Transact-SQL): Returns a value formatted with the specified format and optional culture.

To know more about each of these new functions you can click on the names of the functions.  In my future posts I’ll try explain each of these functions in details with some examples and situation where they can be useful.

 

To know more about the new features introduced in the SQL Server please follow below links:

What's New in SQL Server 2012
What's New in SQL Server 2014

A Sneak Peek into the Future: What's New in SQL Server 2016

Thanks
Mangal Pardeshi

Saturday, July 27, 2013

Different methods to execute large query in small batches

Today I’ll show you some different methods to run a DML operation like UPDATE in small batches for a large table.  Though I’m going to show only UPDATE statements, methods are applicable to DELETE and INSERTS as well.   When you want  to run an UPDATE  query(or other DML operations) over a very huge table having  over a billion rows or even  100s of Million it is recommended that you should not update them at one go.  DML queries on a very large number of rows can cause performance issues, transaction log might get full, affects concurrent users, eats up lot of server resources, runs for hours and many other.   Good practice is that you should break the number of records in small batches of few thousand and update them.  This way you will use minimum server resources and most importantly you will prevent the transaction log file from getting full.

First lets create one table and insert some sample data.

--------------------------------------------------------------------------------------------------------

-- Create the table
CREATE TABLE #A
(
ID INT,
CREATED_DATE  DATE,
METHOD VARCHAR(10)
)

-- Insert 10000 rows in the table
;WITH MANGAL AS
(
SELECT 1 AS ID, CONVERT(DATE,'20000101') AS CREATED_DATE
UNION ALL
SELECT ID+1, DATEADD(DD, 1, CREATED_DATE)
FROM MANGAL
WHERE ID <= 9999

)
INSERT INTO #A(ID, CREATED_DATE)
SELECT ID, CREATED_DATE
FROM MANGAL

OPTION (MAXRECURSION 10000);

-- Check sample data
SELECT *
FROM #A

--------------------------------------------------------------------------------------------------------

This is not the most efficient way to generate some random data using a recursive CTE but I like it so I’m using it.

I. Using a TOP and GO:

This is the simplest method to run a query in a small batches.  I frequently use this method in development when I want to update some records quickly without thinking much.  Note the GO 10 in the query.  An integer after GO will execute the preceding query specified number of times.  In this case update statement will execute 10 times(I know there are 10000 rows in the table and I’m using TOP 1000, simple math).

------------------------------------------------------------------------------------------------------
UPDATE TOP (1000) #A
SET METHOD = 'I'
WHERE METHOD IS NULL
GO 10

-- Verify result
SELECT *
FROM #A

------------------------------------------------------------------------------------------------------

All the queries are quite simple  and they are self explanatory so I’m not going to explain them in detail.

II. Using ROWCOUNT and GO:

Same query but without the TOP operator.  Here I’m using a ROWCOUNT function to limit the number of rows in the batch.  ROWCOUNT causes SQL Server to stop processing the query after the specified number of rows are returned.

------------------------------------------------------------------------------------------------------
SET ROWCOUNT 1000
UPDATE #A
SET METHOD = 'II'
WHERE METHOD <> 'II'
GO 10

SET ROWCOUNT 0

-- Verify result
SELECT *
FROM #A

------------------------------------------------------------------------------------------------------ IMPORTANT:  Using SET ROWCOUNT will not affect DELETE, INSERT, and UPDATE statements in a future release of SQL Server. Avoid using SET ROWCOUNT with DELETE, INSERT, and UPDATE statements in new development work, and plan to modify applications that currently use it. For a similar behavior, use the TOP syntax.

III. Using TOP and GOTO:

Now instead of GO I’ll use GOTO to run the batch multiple times.  The GOTO statement causes the execution of the T-SQL batch to stop processing the following commands to GOTO and processing continues from the label where GOTO points.  Here I’ll use GOTO to keep processing the particular label until it’s @@ROWCOUNT becomes zero.

------------------------------------------------------------------------------------------------------ UPDATE_BATCH:

UPDATE TOP (3000) #A
SET METHOD = 'III'
WHERE METHOD <> 'III'
IF @@ROWCOUNT > 0  GOTO UPDATE_BATCH

-- Verify result
SELECT *
FROM #A

------------------------------------------------------------------------------------------------------

IV. Using ROWCOUNT and GOTO:

Same GOTO concept with ROWCOUNT.

------------------------------------------------------------------------------------------------------

SET ROWCOUNT 3500
UPDATE_BATCH:

UPDATE  #A
SET METHOD = 'IV'
WHERE METHOD <> 'IV'
IF @@ROWCOUNT > 0  GOTO UPDATE_BATCH

SET ROWCOUNT 0
-- verify result
SELECT *
FROM #A

------------------------------------------------------------------------------------------------------

V. Using WHILE and BREAK:

Here I’m taking the help of an infinite WHILE loop ( WHILE 1=1 ) to run the batch multiple times and using the BREAK to exit the loop when @@ROWCOUNT becomes zero.

------------------------------------------------------------------------------------------------------

WHILE (1=1)
BEGIN

    BEGIN TRANSACTION
       
        UPDATE TOP(3333) #A
        SET METHOD = 'V'
        WHERE METHOD <> 'V'
   
        IF @@ROWCOUNT = 0
            BEGIN
            COMMIT TRANSACTION
            BREAK
        END
    COMMIT TRANSACTION
END

-- Verify result
SELECT *
FROM #A

------------------------------------------------------------------------------------------------------

Same can be implemented using ROWCOUNT as I already shown.

VI. Using a Sequence column and WHILE:

Now an entirely different method.  Not using GO, GOTO or BREAK.  This method is more systematic where you have more control on how query is going to execute.  Note that when you use TOP or ROWCOUNT you actually have no control on which rows are going to get updated.  Sorting is completely dependent on the query plan created by the query engine.

------------------------------------------------------------------------------------------------------

DECLARE @C INT
DECLARE @I INT
DECLARE @L INT
DECLARE @MAX INT

SET @C = 1
SET @I = 2200
SET @L = @C + @I
SET @MAX = (SELECT MAX(ID) FROM #A)

SELECT @C AS C, @I AS I, @L AS L, @MAX AS MAX

WHILE @C < @MAX
BEGIN 

    UPDATE #A
    SET METHOD = 'VI'
    WHERE ID >= @C
    AND ID < @L

SET @C = @L -- OR SET @C = @C + @I
SET @L = @L + @I

END

-- Verify result
SELECT *
FROM #A

------------------------------------------------------------------------------------------------------

VII. Using a date range and WHILE:

On similar lines if we don’t have a reliable sequence column or we want to use a date column to update the records by month or year.

------------------------------------------------------------------------------------------------------

DECLARE @C DATE
DECLARE @I DATE
DECLARE @MAX DATE

SELECT    @C = MIN(CREATED_DATE),
        @MAX = MAX(CREATED_DATE)        
FROM #A

SET @I =  DATEADD(YEAR, DATEDIFF(YEAR, 0, @C)+1, 0)

--SELECT @C AS C, @I AS I , @MAX AS MAX

WHILE @C <= @MAX

BEGIN
    UPDATE #A
    SET METHOD = 'VII'
    WHERE CREATED_DATE >= @C
    AND CREATED_DATE < @I

SET @C = @I
SET @I = DATEADD(YEAR, 1, @I)

--SELECT @C, @I
END

-- Verify result
SELECT *
FROM #A

------------------------------------------------------------------------------------------------------

VIII.  Using WHILE without GO, GOTO or BREAK:

One last method from me without using GO, GOTO and BREAK but relatively simpler compare to last 2 methods.  Because different people will advise you not to use GO, GOTO and BREAK in SQL.

------------------------------------------------------------------------------------------------------

DECLARE @ROWCOUNT INT = 1
DECLARE @C INT = 2800

WHILE (@ROWCOUNT > 0)
BEGIN
    UPDATE TOP (@C) #A
    SET METHOD = 'VIII'
    WHERE METHOD <> 'VIII'

SET @ROWCOUNT = @@ROWCOUNT
END

-- Verify result
SELECT *
FROM #A

------------------------------------------------------------------------------------------------------

Same can be done using ROWCOUNT.

IX. Final

Well this is the query I found some 4-5 years back when I was searching for same thing on Google.  I don’t remember where I saw it or who wrote it.

------------------------------------------------------------------------------------------------------
DECLARE @C INT

SET ROWCOUNT 1000
WHILE (1=1)

BEGIN
   BEGIN TRANSACTION
  
   UPDATE #A
   SET METHOD = 'X'
   WHERE METHOD <> 'X'

   SET @C = @@ROWCOUNT

   IF @@Error <> 0
    BEGIN
    Print 'Problem Updating the records'
    ROLLBACK TRANSACTION
    BREAK
    END

   IF @C = 0
   BEGIN
      COMMIT TRANSACTION
      BREAK
   END
   COMMIT TRANSACTION
END
SET ROWCOUNT 1000
-- Verify result
SELECT *
FROM #A

------------------------------------------------------------------------------------------------------

Some notes:  Both TOP and ROWCOUNT can be parameterized.  We can pass a variable instead of direct number.  But of course since ROWCOUNT is soon going to be deprecated from future SQL Server versions one should avoid using it.  I was being lazy not to use transactions (commits and rollback ) wherever possible in above examples but one should use them to add more control, clarity to code and also for better understanding. 

I have not really tested that which query is better, I leave that up to you.  Idea was to share different methods to execute a query in small batches.  But I would go for method 6 or 7 on production environment.  Do let me know your comments, suggestions and what do think of all these methods.

Namaste
Mangal

Sunday, November 14, 2010

ORDER BY Enhancements in SQL Server

Today I am going to talk about one new feature that is been introduced in the SQL Server Denali.   And that is:  one interesting enhancement in ORDER BY clause.  Now with ORDER BY clause you can also specify the OFFSET and FETCH options. 

From Books Online of Denali:
OFFSET: Specifies the number of rows to skip before it starts to return rows from the query expression. The value can be an integer constant or expression that is greater than or equal to zero.

FETCH: Specifies the number of rows to return after the OFFSET clause has been processed. The value can be an integer constant or expression that is greater than or equal to one.

This is how the new syntax of ORDER BY clause looks like in SQL Server Denali:
------------------------------------------------------------------------------------------------------
ORDER BY order_by_expression
    [ COLLATE collation_name ]
    [ ASC | DESC ]
    [ ,...n ]
[ <offset_fetch> ]
<offset_fetch> ::=
{
    OFFSET { integer_constant | offset_row_count_expression } { ROW | ROWS }
    [
      FETCH { FIRST | NEXT } {integer_constant | fetch_row_count_expression } { ROW | ROWS } ONLY
    ]
}

------------------------------------------------------------------------------------------------------

Take a look at following example:

------------------------------------------------------------------------------------------------------
USE AdventureWorks
GO
SELECT
VendorID
    ,Name
FROM Purchasing.Vendor
ORDER BY VendorID OFFSET 5 ROWS
GO

------------------------------------------------------------------------------------------------------

Output:

Offset

In the example I wrote 5 as OFFSET.  You can see in the output 1st five rows are skipped and we got VendorIds starting from 6.  Here SQL Server 1st orders the data on the Column specified in ORDER BY clause(i.e. VendorID).  The next query uses the clause OFFSET 5 ROWS to skip the first 5 rows and return all remaining rows.

Now lets see the example of FETCH:

------------------------------------------------------------------------------------------------------
USE AdventureWorks
GO
SELECT
VendorID
    ,Name
FROM Purchasing.Vendor
ORDER BY VendorID OFFSET 5 ROWS
    FETCH NEXT
3 ROWS ONLY
GO

------------------------------------------------------------------------------------------------------

Output:

FETCH

As you can SQL Server has fetched only 3 rows and that also after skipping 1st five rows.  That is because I specified 5 as OFFSET and 3 as FETCH NEXT.  Here FETCH NEXT 3 ROWS ONLY to limit the rows returned to 3 rows from the sorted result set.

Also:
1. offset_row_count_expression can be a variable, parameter, or constant scalar subquery. When a subquery is used, it cannot reference any columns defined in the outer query scope. That is, it cannot be correlated with the outer query.

2. ROW and ROWS are synonyms and are provided for ANSI compatibility.

3. In query execution plans, the offset row count value is displayed in the Offset attribute of the TOP query operator.

You can find all this information and more on: ORDER BY Clause (Transact-SQL)

Thanks
Mangal

Wednesday, September 1, 2010

Some Things About VIEWS

Everybody has different views about VIEWS, that’s what make them interesting topic to discuss.  Other thing is if you make some assumptions about VIEWS, they can lead you to problems.  As my target audience is SQL beginners, today I’ll talk about few things about VIEWS so some of the obvious mistakes can be avoided.   Actually there are so many things we can talk about VIEWS, but I leave them for future posts.

Today I’ll show what happens when you create a view using “SELECT * FROM TableName” and then ALTER the underlying table used in the VIEW.   The normal assumption most of the SQL beginners make is:  If you create a VIEW using “SELECT * FROM TableName”, all the changes made in the underlying table will automatically reflect in the VIEW as well.

Case 1: You add a column to the table. 
Now here is a question for you:  If you have a table named Employees and a VIEW created  top of it with simple query “SELECT * FROM Employees”.  Now if I add one column to Employees table, will that column appear in result if I execute query “SELECT * FROM View”?

I had ask this question many times in interviews; 80% of time I heard a thumping “Yes”.  And most of the candidate were having experience of well over 3 years.  Answer to above question is BIG NO.  Let me explain this with actual example, 1st lets create some sample data.

------------------------------------------------------------------------------------------------------

-- Create the table with 4 columns
CREATE TABLE
Employees
(
EmpID INT,
FirstName VARCHAR(6),
LastName VARCHAR(20),
ManagerID INT

)
GO

-- Populate the table with sample data
INSERT INTO Employees SELECT
1, 'Mangal', 'Pardeshi', 0 UNION ALL SELECT
2, 'Rahul', 'Sharma', 1 UNION ALL SELECT
3, 'Ajay', 'Varma',2
GO

-- Create a View with wild card *
CREATE VIEW Emp AS
SELECT *
FROM Employees
GO

-- Verify the data in the table and view
SELECT
*
FROM Employees
GO
SELECT *
FROM Emp
GO

------------------------------------------------------------------------------------------------------

Data in the table and view:
test data

As you can see the data  in the Table and View is identical, so are the number of columns.  Now lets add one column HireDate in the table and see what happens.

------------------------------------------------------------------------------------------------------
-- Add HireDate Column in Employees Table
ALTER TABLE Employees ADD HireDate DATETIME
GO

-- Update the HireDate column with some date
UPDATE Employees
SET HireDate = '20100827'
GO

-- Again Verify the data in table and View
SELECT *
FROM Employees
GO
SELECT *
FROM Emp
GO
------------------------------------------------------------------------------------------------------

add column

As you see HireDate column didn’t appear in VIEW.  So question arises: Why a VIEW doesn't get refresh when I add a column to the table? 
Short answer is when you create the VIEW, the column information/definition of VIEW(metadata of VIEW) gets stored in system tables at the time of creation of VIEW.  And that metadata doesn’t get refresh when you alter the underlying table.  You have to explicitly refresh the metadata of VIEW. 
So next question is: How to refresh the VIEW once you modified the underlying table?
There are 2 ways to refresh the VIEW:
1.  Using the system stored procedure sp_refreshview
From Books Online - sp_refreshview: Updates the metadata for the specified non-schema-bound view. Persistent metadata for a view can become outdated because of changes to the underlying objects upon which the view depends.
Syntax: EXECUTE sp_refreshview ‘viewname’

2. Or by executing ALTER VIEW statement.  When you ALTER the VIEW, SQL Server will pick the latest column definition from underlying table and will update the VIEW metadata.

So in our case we can refresh the VIEW Emp by:

------------------------------------------------------------------------------------------------------
-- 1. using system stored procedure sp_refreshview
EXECUTE sp_refreshview 'Emp'
GO
-- 2. Or by executing the ALTER statement
ALTER VIEW Emp
AS
SELECT
*
FROM Employees
GO

-- Verify the data in View
SELECT *
FROM Emp
GO
------------------------------------------------------------------------------------------------------

Note you don’t need to execute both the queries, either of the above can do the trick for you.  Now you can see the HireDate column in Emp view as well:
view

Case 2: You drop a column from the table.
Similarly when you drop a column from the table, the VIEW definition doesn’t get updated even though you have used wild card “*” in VIEW definition.  Now lets drop the column HireDate from the table and see what happens:
Note: In previous step I refreshed the View after adding the HireDate, so now HireDate is part of Emp view as well.

------------------------------------------------------------------------------------------------------
-- First Verify the data before dropping column
SELECT *
FROM Employees
GO
SELECT *
FROM Emp

-- Drop HireDate column from Table
ALTER TABLE Employees DROP COLUMN Hiredate
GO

-- Verify the data after dropping the column.
SELECT *
FROM Employees
GO
SELECT *
FROM Emp
GO

------------------------------------------------------------------------------------------------------

Now in this case HireDate column will get removed from table but the metadata of the VIEW still have its information stored so you will get the following error on selecting data from VIEW.
Msg 4502, Level 16, State 1, Line 1
View or function 'Emp' has more column names specified than columns defined.

The solution is, again refresh the VIEW definition by executing:

------------------------------------------------------------------------------------------------------
EXECUTE sp_refreshview 'Emp'
GO

------------------------------------------------------------------------------------------------------

Now you won’t get the same error.

Case 3: You drop one or more columns and add equal or more number columns to the table.
This case is even more dangerous as User selecting data from the VIEW can get wrong data under wrong columns and can create confusion.  I won’t explain this in detail, just execute following queries and you will realize what I’m saying.

------------------------------------------------------------------------------------------------------
-- Drop column LastName from table
ALTER TABLE Employees DROP COLUMN LastName
GO
-- Add Column DeptName with default value 'IT'
ALTER TABLE Employees ADD DeptName VARCHAR(10) NOT NULL DEFAULT 'IT'
GO

-- Verify the data
SELECT *
FROM Employees
GO
SELECT *
FROM Emp
GO
------------------------------------------------------------------------------------------------------

Final

Now if you see, the LastName column is not present in the table and you can also see the DeptName column in the table.  And interesting observation with VIEW is, though the data is exactly matching with the table but columns names are not correct.  We again need to refresh the VIEW to correct it.

------------------------------------------------------------------------------------------------------
EXECUTE sp_refreshview 'Emp'
GO

-- Verify the data
SELECT *
FROM Emp
GO
------------------------------------------------------------------------------------------------------

Back to normal now:
Last

What is the solution? 
Obvious prevention is don’t use wild card “*” while creating VIEWS.  But even listing out columns is just a prevention or I’d say it is a good practice.   Because even after listing out the columns, if you drop a column from the table that has been used in any VIEW you will still face problems.

The solution is creating the view using “WITH SCHEMABINDING” option.
From Books Online:  Binds the view to the schema of the underlying table or tables. When SCHEMABINDING is specified, the base table or tables cannot be modified in a way that would affect the view definition. The view definition itself must first be modified or dropped to remove dependencies on the table that is to be modified. When you use SCHEMABINDING, the select_statement must include the two-part names (schema.object) of tables, views, or user-defined functions that are referenced. All referenced objects must be in the same database.

Views or tables that participate in a view created with the SCHEMABINDING clause cannot be dropped unless that view is dropped or changed so that it no longer has schema binding. Otherwise, the Database Engine raises an error. Also, executing ALTER TABLE statements on tables that participate in views that have schema binding fail when these statements affect the view definition.

Let me show you one example with SCHEMABINDING:

------------------------------------------------------------------------------------------------------
-- Create View with SCHEMABINDING
CREATE VIEW NewEmp
WITH SCHEMABINDING
AS
SELECT
EmpID,
    FirstName
FROM dbo.Employees
GO

-- Verify the data
SELECT *
FROM NewEmp
GO

-- Now try to alter the table
ALTER TABLE Employees DROP COLUMN EmpID
GO
------------------------------------------------------------------------------------------------------

Now you will get the error:
Msg 5074, Level 16, State 1, Line 2
The object 'NewEmp' is dependent on column 'EmpID'.
Msg 4922, Level 16, State 9, Line 2
ALTER TABLE DROP COLUMN EmpID failed because one or more objects access this column.

Basically  WITH SCHEMABINDING has prevented the change that would affect the view definition.

Additional Information: 
A] If you want see the all the dependant objects on particular table you can use following script:

------------------------------------------------------------------------------------------------------
SELECT *

FROM sys.objects AS so
INNER JOIN sys.sql_expression_dependencies AS sed
    ON so.object_id = sed.referencing_id
WHERE so.type = 'V' AND sed.referenced_id = OBJECT_ID('dbo.employees')

GO
------------------------------------------------------------------------------------------------------

B] The following example creates a script that refreshes the metadata for all views that have a dependency on table dbo.employees:

------------------------------------------------------------------------------------------------------
SELECT DISTINCT 'EXEC sp_refreshview ''' + name + ''''
FROM sys.objects AS so
INNER JOIN sys.sql_expression_dependencies AS sed
    ON so.object_id = sed.referencing_id
WHERE so.type = 'V' AND sed.referenced_id = OBJECT_ID('dbo.employees')
GO

------------------------------------------------------------------------------------------------------

Source:
CREATE VIEW (Transact-SQL)
sp_refreshview (Transact-SQL)

Namaste!
- Mangal

Tuesday, August 24, 2010

Convert Seconds to HH:MM:SS

Today 2 of my colleagues from reporting team had this requirement. They had one table where Seconds were stored as INT and in report they wanted to convert the seconds to HH:MM:SS format.  They already had 1 solution ready with them.   It was something like:

---------------------------------------------------------------------------------------------------------------

DECLARE @Seconds INT
SET @Seconds = 3800

SELECT CONVERT(VARCHAR(10),@Seconds/3600) 
    +':'
    + RIGHT('00'+CONVERT(VARCHAR(2),(@Seconds%3600)/60),2)
    +':'
    + RIGHT('00'+CONVERT(VARCHAR(2),@Seconds%60),2) AS [HH:MM:SS]
  GO

----------------------------------------------------------------------------------------------------------------- The output--
01:03:20

It was working fine, but my colleagues were looking for something different, something elegant.  And I jumped on to help them.  Now a days I hardly get any chance to write SQL, so i don’t let such opportunities go.  I had one solution in mind using CONVERT and Style 108:

--------------------------------------------------------------------------------------------------------------- DECLARE @Seconds INT
SET @Seconds = 3800
SELECT CONVERT(VARCHAR(8), DATEADD(SS,@Seconds,0),108) AS [HH:MM:SS]
GO

---------------------------------------------------------------------------------------------------------------

But problem with the above query is, it fails when number of seconds are more than 86399(there are total 86400 seconds in a day).  So if number of seconds are 86400; above query will show 00 hours instead of showing 24 hours.  See following example:

--------------------------------------------------------------------------------------------------------------- DECLARE @Seconds INT
SET @Seconds = 86500
SELECT CONVERT(VARCHAR(8), DATEADD(SS,@Seconds,0),108)  AS [HH:MM:SS]
GO

---------------------------------------------------------------------------------------------------------------

-- The output--
00:01:40

Here the expected output was 24:01:40.  So I modified the above query a little and I came up with this:

---------------------------------------------------------------------------------------------------------------DECLARE @Seconds INT,@Hour VARCHAR(10)
SET @Seconds = 86500
SET @Hour = DATEDIFF(HH, 0, DATEADD(SS,@Seconds,0))
SELECT STUFF(CONVERT(VARCHAR(8), DATEADD(SS,@Seconds,0),108),1,2,@Hour)  AS [HH:MM:SS]
GO

---------------------------------------------------------------------------------------------------------------

-- The output--
24:01:40

Just in case you want to run above query on table:

--------------------------------------------------------------------------------------------------------------- SELECT STUFF(CONVERT(VARCHAR(8), DATEADD(SS, Seconds,0),108),1,2, DATEDIFF(HH, 0, DATEADD(SS, Seconds,0)))  AS [HH:MM:SS]
FROM TableName
GO

---------------------------------------------------------------------------------------------------------------

I don’t know which approach is better, rather you tell me which one you liked.  Looking at them I think both the queries will give almost identical performance, just that 1st query looks like a Mathematical solution while my approach looks like a SQL solution.  If you know any other approach please feel free to share.

Namaste!
- Mangal